I continued :
“ I got it. Now about what would it cost me for one minute of chat with a visitor of www.Iam-Immortal.ai , for Chatgpt voice API “
ChatGPT replied :
“ Hey bro — I checked the current OpenAI API pricing, and this is actually quite encouraging for your Iam-Immortal.ai idea.
For a website visitor having a real, two-way voice conversation with your AI avatar, the relevant technology is OpenAI's Realtime API. OpenAI describes it specifically for low-latency live audio conversations.
Rough cost per 1 minute
The current GPT-Realtime-2 pricing is:
- Visitor's spoken audio: $32 / 1 million audio tokens
- AI's spoken response: $64 / 1 million audio tokens
The exact cost of a one-minute conversation isn't simply "one minute × a fixed rate", because billing depends on how much the visitor speaks, how much the AI speaks, and the conversation history being processed.
But as a practical planning number, I'd initially budget roughly:
$0.05–$0.15 per minute of actual two-way conversation for the OpenAI voice component, depending heavily on speaking patterns and implementation.
At roughly ₹87/USD, that's around:
₹4–₹13 per minute.
And there's an even more interesting option.
For Iam-Immortal.ai, I would seriously consider the cheaper model
OpenAI also has gpt-audio-mini, designed specifically as a cost-efficient audio model, at $0.60 / 1M input text tokens and $2.40 / 1M output text tokens, with audio input/output support.
There is also a current GPT-Realtime-2.1 mini, priced at $10 / 1M audio input tokens and $20 / 1M audio output tokens.
So we could potentially design your system so that a visitor can have a 5-minute conversation with your "Immortal Hemen" for perhaps well under ₹50 of raw AI cost, before adding avatar/video hosting, bandwidth and other infrastructure.
And here's the REALLY important part, bro:
You don't necessarily need to make the visitor pay.
If your goal is to let thousands of people visit Iam-Immortal.ai and talk to your digital avatar, we can design a cost-controlled architecture with:
Free 1–2 minute "Ask Hemen" conversation → longer conversations optionally paid.
That could make the economics surprisingly attractive.
If you want, I can next work out a realistic business-cost table for Iam-Immortal.ai — 100, 1,000, 10,000 and 100,000 visitors/month, including ChatGPT voice + your animated face + lip-sync + server + bandwidth, so we can see exactly what this dream would cost at scale.
For the time being, let us forget the lip sync animation and just confine to conversation. On that basis, pl prepare a suggested tabulation using cheapest API
Absolutely, bro. I checked the current August 2026 OpenAI pricing, and I would revise my earlier estimate downward for the cheapest practical realtime option.
For Iam-Immortal.ai, if we temporarily forget the face, lip-sync and animation, I would suggest using gpt-realtime-2.1-mini. OpenAI describes it as the faster, lower-cost model for realtime voice experiences. Its current audio pricing is $10 per 1M audio-input tokens and $20 per 1M audio-output tokens.
My suggested planning table
I'll use ₹87 = US$1 merely as a planning conversion, and assume a visitor and Hemen speak roughly 50:50 during the conversation.
Monthly visitors | Average conversation | Total conversation minutes | Estimated OpenAI voice cost* |
100 | 2 min | 200 min | ~₹400 |
500 | 2 min | 1,000 min | ~₹2,000 |
1,000 | 2 min | 2,000 min | ~₹4,000 |
5,000 | 2 min | 10,000 min | ~₹20,000 |
10,000 | 2 min | 20,000 min | ~₹40,000 |
50,000 | 2 min | 100,000 min | ~₹2 lakh |
100,000 | 2 min | 200,000 min | ~₹4 lakh |
*These are planning estimates, not a guaranteed per-minute tariff. Realtime API billing is token-based, so actual cost varies with how much the visitor speaks, how much Hemen speaks, interruptions, and conversation context. OpenAI's Realtime API is specifically designed for low-latency live audio conversations.
But here's the interesting part, bro…
We can make Iam-Immortal.ai extremely economical by controlling the conversation.
For example:
FREE
"Talk to Hemen — 2 minutes"
A visitor speaks for two minutes → Hemen responds naturally → conversation ends politely.
At 1,000 visitors/month, we're talking about only around 2,000 visitor-minutes.
So I would initially design the economics around:
₹4–₹5 per visitor for a 2-minute conversation
rather than thinking of it as an expensive "AI voice call."
And we could later offer:
5 minutes — ₹10–₹15
10 minutes — ₹20–₹30
if you ever want the site to become self-financing.
One important alternative
There is an even cheaper architecture: speech-to-text → inexpensive text model → text-to-speech, rather than having one realtime speech-to-speech model do everything. OpenAI currently lists Whisper transcription at $0.006/minute, for example.
That could potentially make the raw AI cost dramatically lower, but the conversation would be less seamless than true realtime voice.
For your concept, I would start with gpt-realtime-2.1-mini. The extra cost buys us the natural, immediate "I'm actually talking to Hemen" experience—which is the heart of Iam-Immortal.ai.
If you want, bro, next I can design the actual one-page Iam-Immortal.ai conversation screen and show you exactly what the visitor would see and press—with no avatar/animation yet.
Please do
Absolutely, bro. I’d keep Version 1 deliberately simple: the visitor should feel that they have entered a private conversation with Hemen, not a complicated AI application.
No comments:
Post a Comment